Analyzing Complex Predicates: Syntactic Evidences from Sampark and Google Machine Translation Systems

Abstract

The development of a machine translation (MT) system is an extremely challenging task. Because human languages are made up of both simple and complex words. Some words are differently translated depending on their contextual meaning. Therefore, this is of high importance to consider world knowledge for an MT to be able to perform the task of translation effectively. In this sense, MT also produces errors while performing the task of translation. Consequently, measuring such errors of MT output helps us to identify types of errors and subsequently enable us to improve its translation output efficacy. Hence, this study attempts to compare the errors produced by Sampark and Google translators with reference to complex predicates (CPs) while translating it from Hindi to Urdu language. Further, this paper is limited in nature in the sense that it examines Hindi CPs output in Urdu language only. The data for this study have been gathered from different secondary sources like e-dictionaries, e-newspapers, and social media platforms. Next, five hundred Hindi CPs were identified manually and translated using Sampark and Google Translator platforms and errors in their TL outputs were compared. To do so, the CPs were fed to the machine and seen how the machine maps its equivalent translational in the target language i.e. Urdu. On the basis of existing errors in these CP outputs they were divided into five groups i.e. missing word, word order, incorrect word, unknown word and wrong noun-verb agreement and illustrated with examples. Finally, source of the errors have been outlined including their possible reasons.